What is the site fidelity of a dead fish? And how would you prove it?

Animal-bourne loggers and transmitters are often deployed to answer questions regarding animal habitat use. Animal-bourne tags are a powerful tool appropriate to the scale and lifestyle of a large variety of species and are often the only tool available for tracking aquatic, unobservable species. Yet the data derived from tags have important limitations. For example, inferences on the habitat preferences of fish rely on correlations; frequent observations of animal presence in a given habitat are interpreted as evidence of habitat preference. But tagging data are representations of locations in space and time, not measures of movement per se, or even behavior more generally, and these representations can reflect limitations in the sampling technology. Given this reality, how different from a typical tagging result would observations have to be to reject a hypothesis predicting an association between a fish and its habitat? Here we present an approach to developing a null model for animal movement based on random walk simulations, and examine our ability to distinguish random from intentioned movements when using passive acoustics. We compare simulated data to real telemetry observations and provide guidance for studies designed to evaluate habitat preference from passive acoustic telemetry data. We find that the ratio of organismal step size to detection radius interacts with study duration to critically impact calculations of site fidelity; real-world telemetry observations cannot be distinguished from random walks during initial portions of sampling. These studies may require a “burn-in” time before inferences can be fairly made.